r(109) keep if type mismatch

[alessandro alfieri]

hello everybody, I am new with Stata and I have a problem I am not able to solve also with the help of your precious forum.

I want to regress 2 variables: one about age (s2a_03) and another about the emplyment status (ocupado). Below the variables description:

storage display

value

variable

name type format

label

variable label

s2a_03

byte %8.0g

S2 3. �Cu�ntos

a�os

cumplidos

tiene?SI

TIENE

MENOS

DE

1

A�O

ANOTE

00.SI

TIENE

98

	storage	display	value
variable	name	type	format	label	variable label
ocupado	byte	%8.0g	OCUPADO	Poblaci�n Ocupada

This last variable is controversial for me becasue it is displayed to be in a numeric format, but the options are dummy: YES (Si) or NO (NO)

Now, the problem is that I need a subset with 2 criteria:
1. people aged higher than 7 and lower than 13 years old
2. the status of employment as YES (ocupado="Si")

No problems for the first one, as I run:
keep if (s2a_03>=7)
keep if (s2a_03<13)

Lot of problems with the second one, as I run:

keep if ocupado=="Si"

Result is: type mismatch r(109)

I have tried with a lot of other option (drop if, encode...). But I am getting crazy. Please if you can help me, it would be great.

Thank you in advance

[Nick Cox]

ocupado is numeric, not string. So you can't test ocupado for equality with one of its value labels like that. Consider this as a parallel problem.

. sysuse auto, clear
(1978 Automobile Data)

. keep if foreign == "Foreign"
type mismatch
r(109);

. keep if foreign == "Foreign":origin
(52 observations deleted)

So, using this syntax you can go (I guess)

keep if ocupado=="Si":OCCUPADO

What most experienced Stata users would do here is different. Assuming that you coded 1 for Si and 0 for No you could just go something like

regress s2a_03 occupado if inrange(s2a_03, 7, 12) & occupado == 1

except that now all that would do is calculate the mean age in question. A constant predictor is no predictor.

That's a different question. The main point is that there is no need to throw out observations you don't want to use.

[Carlo Lazzaro]

Alessandro:
welcome to this forum.
Your -ocupado- numeric variable is -label-led.
Just type

label list Poblacion Ocupada

from within Stata and the -label-led numeric values will appear.
Then you can do something like:

keep if ocupado==1*assuming that 1 identifies employed people*

to reduce your dataset to the observations you're interested in.

As an aside, it's probably better not to delete observations (as they might be useful for future, currently unexpected analyses), but to flag them with a categorical variable to be used as an -if- qualifier for further analyses.
In your case, you may want to consider:

gen flag=1 if s2a_03>=7 & s2a_03<13 *amended as per Nick's comment (see #4) on my previous mistake*
replace flag=0 if flag==.
regress s2a_03 ocupado if flag==1 & ocupado==1

PS: Crossed in the cyberspace with Nick's reply; his codes are, as usual, much more efficient than mine!

[Nick Cox]

Carlo:

 
 gen flag=1 if s2a_03>=7|s2a_03<13

isn't what is wanted as all values qualify as less than 13 or more than 6. You need & not |.

[Carlo Lazzaro]

Nick is correct:
his code are also much more effective than mine!

[alessandro alfieri]

thank you so much Nick!

:OCUPADO now works for my subset (I can list the variable and see all variable in "Si")

However, running the regression now I have no result that I was expected to find. Thanks to an analysis in absolute terms, I now that the average age is increasing in presence of the status of employed. However, neither with regress, nor with correlate commands I can demonstrate it.

Source	SS	df	MS Number of obs =	372
	F(0, 371) =	0.00
Model	0	0	. Prob > F =	.
Residual	924.731183	371	2.49253688 R-squared =	0.0000
	Adj R-squared =	0.0000
Total	924.731183	371	2.49253688 Root MSE =	1.5788
s2a_03	Coef.	Std. Err.	t P>t [95% Conf.	Interval]
ocupado	0	(omitted)
_cons	10.13978	.0818558	123.87 0.000 9.978825	10.30074

ocupado	s2a_03
ocupado	.
s2a_03	. 1.0000

Do you have some hints?

[Carlo Lazzaro]

Alessandro:
your post has severe formatting issues that make it unreadable as far as your regression model is concerned.
Please share what you typed and what Stata gave you back via CODE delimiters (see the FAQ on this). Thanks.

[alessandro alfieri]

Sorry Carlo, I am gonna try

 
Source
SS
df
MS Number of obs =
372


F(0, 371) =
0.00

Model
0
0
. Prob > F =
.

Residual
924.731183
371
2.49253688 R-squared =
0.0000


Adj R-squared =
0.0000

Total
924.731183
371
2.49253688 Root MSE =
1.5788





s2a_03
Coef.
Std. Err.
t P>t [95% Conf.
Interval]




ocupado
0
(omitted)

_cons
10.13978
.0818558
123.87 0.000 9.978825
10.30074

[CODE][ | ocupado s2a_03
-------------+------------------
ocupado | .
s2a_03 | . 1.0000
/CODE]

[alessandro alfieri]

correlate output:

             |  ocupado   s2a_03
-------------+------------------
     ocupado |        .
      s2a_03 |        .   1.0000

[Carlo Lazzaro]

Alessandro:
are all your variables in numeric format now?

[alessandro alfieri]

Yes Carlo, below the details:

 
describe
ocupado


storage
display
value

variable
name type
format
label
variable label





ocupado
byte
%8.0g
OCUPADO
Poblaci�n Ocupad

 
describe
s2a_03


storage
display
value

variable
name type
format
label
variable label










s2a_03
byte
%8.0g
S2 3. �Cu�ntos
a�os
cumplidos
tiene?SI
TIENE
MENOS
DE
1
A�O
ANOTE
00.SI
TI

[Nick Cox]

You still have problems with formatting output (please keep reading FAQ Advice #12 and don't try variations of your own) but the problem reported in #6 and #8 and #9 is just the problem warned about in #2. Your predictor has just one distinct value. You threw out all the observations for the other value. So, your set-up is analogous to trying to fit a regression to data like this:

sysuse auto, clear 
scatter mpg foreign if foreign, xla(1, valuelabel)

So, remember that regression with one predictor means that you are trying to fit a straight line. That's easy: the straight line has intercept the mean of the response or outcome and slope zero. Regression makes perfect sense, but it's equally true that the predictor has no predictive value.

Further, the correlation is indeterminate because the variance of one variable, one term in the denominator, is precisely zero.

At a minimum you need both values for the predictor and perhaps other predictors too.